Packaging automation drives WDM component assembly

Wdm93791 37

Scott C. Jordan

The similarities between the recent fiberoptic industry hypergrowth phase and the pattern of earlier high-tech markets are sobering. Boom-and-bust cycles have characterized markets as diverse as semiconductor fabrication and disk-drive manufacturing, driving waves of consolidation and merger on the financial side. A common thread is seen: in order to gain control of oscillations in the supply/demand relationship, both a production-automation ethic and a relentless continuous-improvement paradigm emerged.

In particular, the semiconductor industry was an early adopter of sophisticated process automation and throughput-optimization (and more recently, utilization-optimization) tools to drive costs down and yields up. This adaptation has allowed that industry better control of its economic destiny.

The lesson has not gone unnoticed here in the fiberoptic industry. Consequently it has become increasingly accepted in WDM component manufacturing that similar levels of process automation must occur here if we are to avoid a punishing bust cycle when supply eventually meets demand.

The largest cost in packaged WDM devices is the cost of packaging. For example, a laser diode costs a few dollars, but a pigtailed laser diode costs hundreds of dollars. Contributing to this enormous increment are yield and throughput issues inherent in the packaging process, which has remained a manual one for many manufacturers.

Emergence of packaging automation

Numerous instruments and systems have come to market to automate the exacting positioning of fiberoptic devices in packaging operations. It is this alignment step that is so costly and prone to yield headaches (see Table 1).

The oldest automated-alignment approach is based on an analog phase demodulation technique utilizing a continuous looping dither of one optical element (such as a single-mode fiber) versus the other (for example, a laser diode). An error signal derived from the phase of the observed optical coupling signal versus the dither rapidly drives the system towards optimum alignment.

A host of devices based on this technique have come to market, most based on piezoelectric motion stages, plus one memorable unit that used a voice-coil mechanism to motivate a small steering lens for optimizing coupling to a free-space laser beam. Automated alignment systems based on analog phase demodulation are compact and align very quickly in "clean" single-mode couplings but share the drawback of potentially locking on to local maxima, which results in untidy coupling.

The field of industrial fiberoptic packaging automation progressed gradually following introduction of the first successful stack-of-stages system, devised by the author in 1989. This system used a new digital gradient search algorithm. The attraction of a digital gradient search is that it leveraged the limited amount of data that can be acquired in a practical amount of time. This feature allowed the practical use of large crossed-roller bearing stages, providing a significant advance in industrial compatibility compared to systems previously available.

Systems based on a digital gradient search were more costly and bulky than the analog units, but they could provide long travels and it was a straightforward matter to program additional search and scan algorithms to accommodate applications requirements or to sidestep the problem of local-lock-on in the gradient search process. These systems were eventually extended to integrate bonding functionality, pick-and-place mechanisms, video imaging, and other capabilities, making them ideal for applications where a high degree of integration by the supplier was desired. Today several suppliers provide stacked-stage-based alignment subsystems suitable for integration into such systems.

With the commercial advent of thin-film WDM components and MEMS switches and cross-connects in the past several years, however, flexible angular alignment automation is also required. These components use collimated input and output fibers, for which angular alignment is typically much more critical than transverse alignment.

A hexapod-based microrobot is one solution to this requirement. The microrobot integrates long-travel six-degree of freedom motion and angular as well as transverse alignment automation. The hexapod configuration-best described as a miniaturized version of a flight simulator-and digital controls allow real-time automatic coordinate transformation, in turn allowing the center of rotation to be instantly placed anywhere in space through firmware. Rotations can thereby be commanded to center at the tip of a cleaved fiber, the beam waist of a laser diode, the focal point of a lens, the surface of a thin-film filter, the optical axis of an array channel, or any other desired point in space (see Fig. 1).

The microrobot is an enabling technology for efficient submicron packaging automation of even the newest class of array and MEMS devices and collimated elements. In applications requiring nanometer-scale resolutions, a closed-loop XYZ piezo module can be added. In addition to the microrobot`s native auto-alignment algorithms, the high speed and resolution of this approach allows the full field of a coupling cross-section to be mapped in less than three seconds, eliminating spurious lock-ons to local coupling maxima-a perennial contributor to poor yield for processes based on gradient search techniques.

Economics of packaging automation

To understand the economic advantages of automating the packaging process, consider the case of an alignment process where a manually-configured interconnect requires 15 minutes of technician labor at 40% yield per interconnect. Automating this process will allow the technician to move among several workstations-say three for the purpose of this example-while boosting yield, to (for example) 80%. Assume a burdened labor cost of $25/hour.

The economic benefits of automation are easily seen. In the worst case, interconnect yields are multiplicative, resulting in a cost structure that rises exponentially with the number of interconnects in the package (see Fig. 2).

For the simplest situation-a single interconnect, such as in a pigtailed laser diode-the cost benefits of automation are already substantial, with the cost per interconnect in the automated case below 25 cents compared to more than $15 per interconnect for the manual case. For more complex devices such as DWDM add/drops, switches, and cross-connects, the mounting costs of labor-intensive operations and poor yield quickly become staggering. In reality, of course, use of ribbon arrays would be more efficient for high-channel-count devices, but the point is still clear: manufacturing efficiencies quickly become competitive advantages.

Continuous improvement in fabrication

Our industry is learning the same lesson that the data-storage and semiconductor industries have learned: time is money. Automating the packaging process is key to economic survivability, but there are other areas in the fabrication of fiberoptic devices where money is wasted. In particular, as precisions tighten, the settling time of the assembly equipment after a motion device has been actuated has emerged as the single greatest obstacle to throughput and resolution improvement in fabrication processes.

Consider fiber grating fabrication. Many techniques exist for generating grating structures in fibers, such as using bulk interferometry1 or diffractive phase masks to generate gratings of desired length, inclination, and profile. Mask-based approaches include modulating the position of a diffractive phase grating perpendicular2 or parallel3 to the fiber axis.

Such processes are sensitive to nanometer-scale positioning errors. Not only must the positioning mechanism perform to this degree, but all relative motions in the system must be suppressed commensurately. Isolation tables attenuate ambient disturbances, but structural ringing driven by onboard devices still takes hundreds of milliseconds to damp out-a severe throughput penalty.

A review of vibrational physics is worthwhile here. After a disturbance, the amplitude of the resonant ringing of each element in a structure scales as e-t/τ, where τ is the time constant for each element`s resonant characteristics.4 For structures with damping characteristics typical of precision motion subassemblies, τ ~ (ωnζ)-1 where ωn is the resonant angular frequency and z is the damping ratio for the resonance. The symbol ζ is commonly defined as the ratio of the damping for the resonance versus critical damping [ζ= C/Cc] and varies from 0 (no damping) to 1 (critical damping) (see Table 2 and Fig. 3).

Clearly, this results in worsening process throughputs as tolerances tighten. Furthermore, many phase-mask approaches require moving the grating in continuous dither motions, not stepwise.5

Mechanical resonances thus degrade the fidelity of the generated grating and so its optical performance, reducing efficiency and smearing channel width. Boosting the servo-loop gain in the dither mechanism can result in more accurate trajectories but greatly increase the system`s susceptibility to motion-driven vibration.

Fortunately, classical damping is not the only tool for eliminating structural resonances. A real-time feedforward technology called Input Shaping was developed based on research at the Massachusetts Institute of Technology (Cambridge, MA) and commercialized by Convolve (New York, NY).6 It is now an integrated option for the latest digital piezo controllers.

The technology acts transparently in real time to prevent the motion-driven excitation of resonances throughout the system, including all fixturing and ancillary components. Unlike notch filtering, it is insensitive to variations in each resonant frequency over a range exceeding ±15%, is effective against multiple frequencies and resonances occurring outside the servo loop and even the servo bandwidth, and does not require recalibration of servo settings (see figures 4 and 5).

Boom-and-bust examples in other high-technology industries pose sobering lessons for the fiberoptic industry. The priorities for the industry must center on process automation-particularly for the costliest process: packaging-and throughput optimization. Promising new technologies and tools are emerging that will help optimize process economics and device efficiencies.


1. For example, Glenn et al, US Pat. 4725110; Mizrahi et al, US Pats. 5363239 & 5366304, or Perez & Tyagi, 5709738.

2. For example, Epworth and Bricheno, US Patent 5629603.

3. See, for reference, the excellent overview of long and apodized grating generation in Brennan et al., US Pat. 5912999.

4. A fine reference for the physics of resonant vibrations is Modern Control Engineering by Katsuhiko Ogata (Prentice-Hall, 1970, ref. page 271). A remarkable on-line reference can be found at /spr95/theory/detsys/detsys_1.html.

5. Ibid.

6. See US patent 4,916,635 and 5,638,267.

Scott C. Jordan is the director of nanopositioning technologies at Polytec PI, 1342 Bell Avenue, Suite 3A, Tustin, CA 92780. He can be reached at e-mail: or 714-850-1835.Wdm93791 37Wdm93791 38

FIGURE 1. Compact six-axis industrial microrobot is based on a hexapod principle, allowing six-degree-of-freedom-motions with 100-nm resolution plus rotations about any point in space, such as a fiber tip, laser beam waist, lens focal point, thin-film filter surface, or array channel axis.Wdm93791 39

FIGURE 2. The cost due to alignment varies for packages with a different number of interconnects. The increase in yield and decrease in technician time per interconnect is highly leveraged, especially for more complex devices with more interconnects. Wdm93791 40

FIGURE 3. Generic mechanical damping behavior after a rapid motion, shown for various damping ratios, z. The amplitude of the ringing diminishes with time as e-t/t, where t ~ (wn z)-1.Wdm93791 41Wdm93791 42

FIGURE 4. Quasi-sinusoidal phase-mask dither waveform is corrupted by fixturing resonances excited by stop-start and timing-glitch transients and other non-sinusoidal components of the motion profile (a). The same input waveform with structural resonances nullified shows the benefit is even more impressive for harsher motions such as square-waves (b). Elimination of structural resonances results in higher-fidelity Bragg and transmissive fiber gratings: narrower channel-widths, higher efficiency.Wdm93791 43

FIGURE 5. In a coarse/fine air-bearing/piezo stage, the closed-loop piezo stage is used for high-throughput nm-scale mask position modulation in grating manufacture. Vibration-nullification technology allows higher production throughput while optimizing grating fidelity for narrower channel-width and improved efficiency.

More in DWDM & ROADM